(088) Capturing Information Not Presented in Hazard Ratios: The Added Value of Presenting Plots of Instantaneous Hazards for Time-To-Event Analyses

Background: Longitudinal studies in pharmacoepidemiology typically report hazard ratios (HRs) as the summary measure of time-to-event analyses. HRs are limited in that they only convey the relative ratio of cumulative hazards between exposure groups averaged over the entire study period. They are not useful for answering certain clinically important questions related to the timing of a treatment’s highest or waning effectiveness.

Objectives: To demonstrate the added value of instantaneous hazards for survival analyses in the context of a comparative effectiveness study of monoclonal antibody (mAb) pre-exposure prophylaxis (PrEP) against COVID-19 hospitalisation among the immunocompromised population.

Methods: We conducted a retrospective cohort study harnessing Clalit Health Services electronic medical records data in Israel to identify patients who received mAb PrEP and matched them on propensity scores (PS) to unexposed individuals (i.e., those who did not receive mAb PrEP). Cox proportional hazards regression models were used to estimate HRs and 95% confidence intervals (CIs) for the effectiveness of mAb PrEP against COVID-19 hospitalisation. We also estimated and plotted non-parametric smoothed estimates of the instantaneous hazards for both exposure groups.

Results: The study population comprised 2444 mAb-exposed patients matched to 2444 unexposed patients. In the 6-month follow-up, the estimated unadjusted HR (95% CI) was 0.68 (0.43–1.08) in the PS-matched analyses. Plots of instantaneous hazards showed that mAb exposure was associated with a lower rate of COVID-19 hospitalisation until Day 120 of follow-up, after which, the upper 95% CI of the ratio crossed 1. It was also evident from the plots of the instantaneous hazards that waning of mAb effectiveness began at around Day 80 of follow-up and, by Day 180, both exposure groups were experiencing the event of interest at a similar rate.

Conclusions: While the overall HR suggested a lack of statistically significant effectiveness across time, the instantaneous hazards plot indicates that effectiveness plateaued at Day 80 and then waned over time. A limitation of instantaneous hazards is that calendar time is not taken into account, and selection bias can be introduced due to informative censoring. The latter can be addressed by inverse probability weights. Plots of instantaneous hazards can help convey important information from time-to-event analyses that are not captured by overall HRs, including measures of absolute risk. Consideration must be given to presenting such plots to complement overall HRs in time-to-event analyses.